Data Center Strategy

Data centers are where the AI boom becomes physical, and the capital committed is now infrastructure-scale. The four largest hyperscalers (Microsoft, Amazon, Alphabet, Meta) spent roughly 240 billion dollars of capex in 2024 and roughly 380 billion dollars in 2025, most of it AI data center capacity (company filings and guidance), with announced megaprojects such as OpenAI's Stargate program carrying headline commitments of up to 500 billion dollars over four years. That spending pulls six distinct value chains at once: colocation landlords, power producers and fuel, electrical gear, cooling, construction, and the servers, storage, and networking that fill the racks. Global data center electricity use was about 415 TWh in 2024, roughly 1.5 percent of world demand, and the IEA's base case projects about 945 TWh by 2030 (IEA, Energy and AI, 2025).

The binding constraint has shifted from silicon to electricity delivered on schedule, and scarcity is now visible in market prices: PJM capacity auctions cleared at 269.92 dollars per MW-day for 2025/26 and 329.17 dollars for 2026/27, versus 28.92 dollars two auctions earlier, and Northern Virginia vacancy is below 1 percent with a record roughly 6.6 GW under construction in North America at the end of 2024, most of it preleased (PJM, CBRE). The structural question for the sector is not whether the buildout is happening. It is whether tenant AI revenue grows into the capital bill before efficiency gains, power constraints, or financing stress force a slowdown, and which layers of the supply chain hold pricing power in either scenario. This page is analysis and education, not investment advice.

The roster sorts into seven archetypes along the supply chain, in the order a capex dollar travels. This is framing, not stock picking.

1. Landlords (Equinix, Digital Realty, Iron Mountain, GDS). REIT economics: long leases, development yields versus cost of capital, and interconnection ecosystems. Retail colo with cross-connect density (Equinix) carries different pricing power than wholesale build-to-suit (Digital Realty) or a records business funding a data center pipeline (Iron Mountain). GDS adds China and Southeast Asia exposure with its own regulatory risk set.

2. Power scarcity owners (Vistra, Constellation, NRG, Talen). Competitive generators whose existing fleets gained a new buyer class. Nuclear PPAs at premium prices (Constellation, Talen) and gas-fleet capacity value (Vistra, NRG) are the cleanest expression of electricity scarcity in the roster.

3. The power island shovel sellers (GE Vernova, EQT). Turbines, grid equipment, and the natural gas that fuels the bridge. Backlog-driven industrial economics (GE Vernova) and commodity production with a structural demand pull (EQT) rather than data center revenue per se.

4. Electrical gear (Eaton, Schneider, ABB, Hubbell, nVent, Powell). Diversified electricals where data centers are the growth engine but not the whole business. The large caps (Eaton, Schneider, ABB) offer scale and backlog visibility; the smaller names (nVent, Powell) carry more torque per data center dollar and more cyclicality.

5. Thermal management (Vertiv, Trane, Johnson Controls, Carrier, Modine). Vertiv is the pure-play reference for both power and cooling inside the white space. The HVAC majors sell applied chiller systems into the same buildout from diversified bases. Modine is the small-cap with the highest data center revenue mix shift. The liquid cooling transition is the segment's key share battle.

6. Builders (EMCOR, Comfort Systems, Quanta, Sterling Infrastructure). Labor-constrained contractors converting the pipeline into structures, power connections, and mechanical systems. Backlog and remaining performance obligations are the forward indicator; skilled-labor supply is the binding constraint on their growth.

7. Rack payload and cluster fabric (Dell, Supermicro, HPE; Western Digital, Seagate, Pure Storage; Arista, Coherent, Lumentum, Fabrinet, Credo, Astera Labs). AI servers are high revenue at thin integration margins. Nearline storage is a structural duopoly (Western Digital, Seagate) with flash (Pure Storage) attacking from above. Networking splits into switching (Arista), optics (Coherent, Lumentum, Fabrinet), and interconnect silicon (Credo, Astera Labs), all levered to cluster scale and 800G-plus transitions.

Cross-cutting axes: purity of exposure (Vertiv, Equinix, Arista versus diversified industrials), position in the demand chain (gear orders lead, construction and payload deliver later), and contract duration (15-year leases and 20-year PPAs versus book-and-ship hardware). The useful question is usually not which name is best, but which layer holds pricing power at the current point in the buildout and how much of that is already capitalized.